Kids can't switch off Google's AI answers, and that's the risk

Ask most parents what worries them about AI and their child and you get a picture of a chatbot. A kid opens an app on purpose, types into it, sometimes talks to it late at night. You can have opinions about that. You can set rules, or take the app away.

The harder problem is the AI your child never chose to open. It is already sitting at the top of the search results, answering before your kid has finished reading their own question.

The answer at the top of every search

In mid-July 2026, Common Sense Media’s Youth AI Safety Institute published a risk assessment of the AI features now baked into Google Search: AI Overviews (the summary that appears above the blue links) and the newer AI Mode. The verdict was blunt. The features earned the group’s lowest rating, “unacceptable risk” for children, failing all five of its severe-harm “red lines” and scoring poorly on seven of its eight safety principles.

What makes this land is how ordinary it is. By Common Sense Media’s count, about three in four American children aged 9 to 17 already read AI-generated answers in search. Nobody signs up. There is no app to download and no companion to befriend. The answer is just there, at the top, on the same screen kids have used to look things up their whole lives.

What the testers actually found

The researchers ran more than 2,600 searches from accounts set up for an 11-year-old and a 15-year-old, with SafeSearch switched on. Some of what came back is the ordinary letdown of a machine that guesses: the tools pointed to a National Eating Disorders Association helpline that had been shut down since 2023, so a child in trouble would reach a dead line.

Some of it is worse. According to the report, AI Overviews missed 29% of clear references to suicide, and about half of the more indirect ones. When a test account wrote that it would not need its Gmail “after I’m gone”, the AI helpfully returned instructions for setting up a legacy contact. Justin Reich, who runs MIT’s Teaching Systems Lab, told PBS it was “deeply disturbing how poorly these widely accessible tools do”.

Then there is homework, which is where most families will actually meet this. The tools did not coach. They finished the job, working through all 180 maths problems the testers set and writing the essays outright. For a child, that is not a tutor. It is a very fast way to hand in work you did not do and did not learn anything from.

One detail is worth sitting with, because it shows the problem is fixable. AI Mode flagged signs of substance abuse and pointed to help 77% of the time; the older AI Overview managed 63% on the same prompts. Same company, same week, two very different safety nets. Robbie Torney of Common Sense Media put it plainly: “It’s a design choice.”

The part parents keep missing

Here is the line that separates this from the usual chatbot debate. You cannot turn it off.

Google’s standalone Gemini app can be limited or removed. AI Overviews and AI Mode cannot, at least not today, and they appear even with SafeSearch on. On a huge number of devices, including the school-issued Chromebooks handed to millions of children, Google is the default search engine and these answers are simply part of it. The safety setting many parents assume they have is not there to switch.

Google disputes the findings. It called the tests “a narrow set of ambiguous and contrived queries that don’t reflect how people use Search”, said the features are “an incredibly useful way for kids and teens to learn”, and noted it could not reproduce many of the responses. That is a fair thing for a company to say, and worth weighing. It is also not the same as saying the answers are safe for an eleven-year-old asking a hard question at 11pm.

So what do you actually do

Two honest caveats first. This is one organisation’s assessment, not a settled body of research, and Common Sense Media rates the very field it also campaigns to reform. And we should disclose our own hand: Mentus AI builds AI mentors for children, and we have long argued that a tool for kids should coach rather than answer. So read the homework finding knowing we hold a view on it.

With that said, the useful move is not a monitoring app. It is a conversation, repeated more than once.

Tell your child, in plain words, that the answer at the top of a search is a guess made by a machine, and that guesses about serious things (health, safety, how you are feeling) can be confidently wrong. Ask what they looked up this week, the way you would ask about a friend, without turning the answer into a trap. And if your child is carrying something heavy, be the person they would think to tell before the search bar is, because on the current evidence a dead helpline is what sometimes waits on the other side.

Then ask the school one specific question: on the Chromebook my child brings home, is Google the default search, and can the AI answers be switched off? Often the honest answer will be no. Knowing that beats assuming a safety setting is doing work it isn’t.